AI Agent Hub
Back to plugins
🖥️

dsh-sandbox-micro

Client Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install omdsh-dev/dsh-sandbox-micro

Paste the following prompt into your AI chat to install this plugin:

Please run the installation command in DeepSeek Harness to enable this plugin, source code address is https://github.com/omdsh-dev/dsh-sandbox-micro.

About this plugin

When handling untrusted commands generated by models, traditional shell execution methods often face security risks related to host environment injection, especially on Windows platforms where parameter forwarding via cmd.exe can easily lead to parsing errors or security vulnerabilities. The dsh-sandbox-micro plugin addresses this isolation issue entirely by introducing a Linux micro-VM based on microsandbox. It runs a built-in Node shim executor directly, rather than relying on a host shell wrapper, ensuring commands are executed in their raw form within a strict isolated environment. This solution not only supports memory limits, timeout controls, and optional network access policies but also provides an impregnable security barrier for code generation and automation tasks through its Fail-Closed mechanism, which refuses execution if prerequisites are not met.

For developers requiring high security standards or dealing with complex workflows, this plugin offers flexible and powerful sandboxing capabilities. It preserves the seamless ctx.sandbox interface while exposing the micro-VM environment to the model via ctx.shell, supporting various strategies ranging from read-only mounting to full isolation. With built-in Debian images and configuration options, rapid deployment and debugging become possible, allowing users to confidently execute automated scripts, code analysis, or build tasks.

Use Cases

  • Safely execute untrusted commands generated by the model
  • Run automated scripts in an isolated Debian environment
  • Verify the security of generated code logic

Best For

  • Developers requiring safe execution of AI-generated code
  • Enterprise architects managing automated workflows
  • Researchers focused on container isolation security